Bob,

Yes, I am starting to appreciate the scope of this problem. As an AFNI user I often wondered how this issue was handled in SPM. Recently I learned it is not really handled there either and that there is no single approach to the hierarchical analysis of fMRI data. Our solution has always been to try to collect enough data so that small violations of underlying assumptions have little effect on the final outcome. In addition we tend to apply the most conservative techniques possible (ie actually taking between subject variability into account with a second level of analysis). Of course this is not always done. From an inference standpoint it makes it very difficult to compare data from different groups since even if it is clear how the data were analyzed, it is remains unclear how such methodological differences affect the specific pattern of results.

Anyway, enough venting for now. It is just nice to hear that we are not the only ones fighting with these issues.